摘要
神经机器翻译为加深世界交流做出了巨大贡献,它的发展促进了世界化的发展。研究针对基础的Transformer模型存在的问题,对Transformer模型进行改进,进而提出一种组合式神经机器翻译模型。该模型引入ELMo、Mix-BA以及DMAL,优化了机器翻译对单词的表达形式、多头注意力层之间的联系以及句子中重点单词的关注度。研究利用WMT14ende数据集与IWSLT14de-en数据集进行对比实验,在两种数据集中,组合式神经机器翻译模型的BLEU得分相较于Transformer基线模型分别高出1.07、0.92;在长句翻译中,组合式神经机器翻译模型的BLEU评分达到33.56,并高出LSTM模型5.72。结果表明研究所提出机器翻译模型具有更好的翻译效果,为神经机器翻译的发展提供新的思路。
Neural machine translation has made great contributions to deepening world communication,and its development has promoted the development of globalization.Aiming at the problems existing in the basic Transformer model,this paper improves the Transformer model,and then proposes a combined neural machine translation model.This model introduces ELMo,Mix-BA and DMAL to optimize the attention of machine translation to the expression form of words,the connection between multiple attention levels and the key words in sentences.The study used WMT14en-de data set and IWSLT14de-en data set to conduct a comparative experiment.In the two data sets,the BLEU score of the combined neural machine translation model is 1.07 and 0.92 higher than that of the Transformer baseline model,respectively;In long sentence translation,the BLEU score of the combined neural machine translation model reached 33.56,which was 5.72 higher than that of the LSTM model.The results show that the machine translation model proposed by the research institute has better translation effect and provides new ideas for the development of neural machine translation.
作者
宫昀
GONG Yun(Xianyang Normal University,Xianyang Shaanxi,712000,China)
出处
《自动化与仪器仪表》
2023年第8期257-261,267,共6页
Automation & Instrumentation
基金
陕西省“十四五”教育科学规划2022年度课题(SGH22Y1419)
2023年度陕西省哲学社会科学研究专项(2023QN0273)
咸阳师范学院、陕西省教育学会2021年教育教学改革研究项目(2021Y034)。
作者简介
宫昀(1981-),女,陕西咸阳人,文学博士,讲师。
